Security and Hardening treats every external input as hostile and starts each feature with a short threat model: map trust boundaries, name assets, apply STRIDE and write abuse cases. A three-tier boundary system lists what to always do, what needs human approval and what never to do.
The skill covers injection, XSS, access control, authentication and sessions, headers and CORS, input validation, uploads, SSRF, destructive operations on derived paths, rate limiting, secrets, dependency audits, personal data and LLM output. Worked code sits in a references file.
What You Get
- A threat modeling process with a STRIDE table
- Always do, ask first and never do boundary lists
- Control guidance for the OWASP classes, SSRF, rate limiting and secrets
- A references file of code patterns, a red flag list and a verification checklist
Common Use Cases
- Handling user input and authentication
- Reviewing a feature that fetches user-supplied URLs
- Auditing dependencies and supply-chain risk
- Adding AI features or collecting personal data
What Makes It Unique
Besides the usual OWASP controls, the skill includes a rule for destructive operations on derived paths, with allowlisted root, minimum depth and ownership evidence, plus dependency steps that block install scripts and judge reachability. It also covers privacy duties and the OWASP list for LLM applications.
Under The Hood
The SKILL.md is the workflow and a references file holds TypeScript patterns. The skill runs no scripts of its own. It tells the agent to run package-manager audits, git diff checks for secrets and install-script policies, which reach package registries, and to read a shared security checklist in the repository.